Although metabolic and physiological imaging have shown prognostic and predictive value for radiation therapy (RT) outcome and may serve to support therapy modification, clinical utilization is challenging due to issues such as reproducibility of physiological images, heterogeneity of the image parameters in the tumor, and lack of tools to support therapy modification. We hypothesize that, given the underlying heterogeneity of tumors and their response to treatment, physiological imaging can be best utilized for individualizing RT by robustly and quantitatively identifying subvolumes that have a better predictive value than the imaging characteristics of the tumor as a whole, and are the likely sites of local failure. These subvolumes can serve spatial guide for radiation dose redistribution such as focal boosting to improve outcomes. We have developed and investigated fuzzy logic pattern recognition techniques for identifying these subvolumes of head-and-neck (HN) cancer from heterogeneous distributions of tumor blood volume (BV) across patients and over multiple time points. Based on our findings that large poorly-perfused subvolumes of HN tumors before treatment that persist during the early course of chemo-RT have the potential to predict local failure better than the change in the mean BV in the tumor, we will further develop the subvolume definition method, extend it to diffusion- weighted (DW) MR imaging, and evaluate and validate it in a randomized phase II clinical trial of poor- prognosis HN cancers. Our aims are: (1) Develop quantitative and automated methods to extract significant subvolumes of HN tumors from dynamic contrast enhanced (DCE) and DW MRI for prediction of local and regional failure; (2) Prospectively assess variability and reproducibility of the subvolume intensity and definition extracted by our methods using test-retest data; and (3) Prospectively evaluate and validate that the method yields subvolumes predictive of local-regional failure in a randomized phase II trial of radiation dose boosting for poor-prognosis HN cancer. Impact: Unless clinics can acquire and analyze these images effectively and quantitatively, and use predictive results in an accurate fashion, the benefit of these powerful imaging techniques to make a meaningful difference in tumor RT may be completely obscured. This project strives to achieve these goals. The framework developed in the project can be translated into other body sites as well as other imaging modalities. By partnering with a commercial company, we will make this method available for the users in the community.